Computer Vision Model

Fashion MNIST Dataset

Created a simple deep neural network for computer vision. Trained to recognize a piece of clothing from the Fashion MNIST Dataset. Note: In this model, the clothing item has to be the only thing on the picture and it has to be centered. Accuracy: Test Data: 83% ; Training Data: 87%

Convolutions And Pooling

• Created a feature map of an image from the from misc in SciPy.

  

• The first filter gives emphasis to the vertical lines of the picture.

  

• The second filter provides emphasis to the horizontal lines of the picture.

  

• Pooling is done using Maximum Pooling. Information from the picture is reduced (in terms of pixels) while maintaining the features that are detected.

  

Enhance Computer Vision Using Convolutions

Enhanced Computer Vision for the MNIST Fashion Dataset using a Convolution Matrix. Accuracy: Test Data: 91% ; Training Data: 99%